Multilevel Thresholding Based on Exponential Cross Entropy and Niche Chaotic Particle Swarm Optimization

Abstract

The existing method of threshold selection based on maximal Shannon entropy or exponential entropy neglects the relationship between object and background to some extent. Aimed at the above-mentioned problem, a threshold selection method based on exponential cross entropy is proposed in this paper. Being different from maximal Shannon entropy only based on histogram distribution, minimum exponential cross entropy minimize the difference between the amount of information of original image and that of segmented image. First, exponential cross entropy is defined and the corresponding method of single threshold selection is derived. Then the method is extended to multilevel threshold selection. Furthermore, the niche chaotic mutation particle swarm optimization algorithm is adopted to find the optimal multi-thresholds. A large number of experimental results show that, compared with the method of multilevel threshold selection based on maximum Shannon entropy with particle swarm optimization (PSO), the segmented images of proposed method are more accurate and their visual effect is improved significantly.